| name | context-budget |
| description | Audit token consumption across agents, skills, MCP servers, and rules -- identify bloat, redundancy, and optimization opportunities. |
| origin | ECC |
Context Budget
Audit token consumption across agents, skills, MCP servers, and rules to identify optimization opportunities.
When to Use
- Context window feels constrained during complex tasks
- After installing many skills, agents, or MCP servers
- Periodic maintenance to keep token overhead low
- When agent performance degrades due to context bloat
How It Works
Phase 1: Inventory
Scan all components and estimate token usage:
- Skills (SKILL.md files)
- Agents (agent definitions)
- MCP servers (tool schemas -- ~500 tokens per tool)
- Rules (rule files)
Phase 2: Classify
Categorize components by usage frequency:
- Always needed: Core skills and rules used in every session
- Sometimes needed: Domain-specific skills activated on demand
- Rarely needed: Specialized tools used infrequently
Phase 3: Detect Issues
Identify problems:
- Bloated descriptions that could be shorter
- Redundant components covering the same ground
- MCP over-subscription (too many tools registered)
- Agent descriptions that load in every Task tool invocation
Phase 4: Report
Generate a breakdown with optimization recommendations ranked by potential token savings.
Key Insights
- MCP servers represent the largest token overhead (~500 tokens per tool schema)
- Agent descriptions load in every Task tool invocation
- Skills only load when activated, making them more efficient
- Rules load on every conversation start
Modes
- Basic audit: Summary of component counts and estimated token usage
- Verbose audit: Detailed per-component breakdown with specific recommendations